WO2007011160A1

WO2007011160A1 - Apparatus and method of embedded quantizaton for the improved snr scalbilty

Info

Publication number: WO2007011160A1
Application number: PCT/KR2006/002829
Authority: WO
Inventors: Hae-Chul Choi; Jae-Gon Kim; Jin-Woo Hong; Sung-Je Ko
Original assignee: Electronics And Telecommunications Research Institute
Priority date: 2005-07-19
Filing date: 2006-07-19
Publication date: 2007-01-25

Abstract

Provided are a method and apparatus for enhancing coding efficiency by performing encoding and decoding by optimally calculating quantization intervals and reconstruction values through a distribution of DCT coefficients of each frame, when DCT coefficients of each SNR enhancement layer are quantized in scalable video coding. The encoding apparatus includes: an R-D optimization unit performing Rate-Distortion (R-D) optimization based on a distribution of Discrete Cosine Transform (DCT) coefficients of each slice and calculating a first reference value and a second reference value respectively indicating a start point and an end point of DCT coefficients quantized to '0'; a quantization interval setting unit setting adaptive quantization intervals on the basis of a minimum value and a maximum value of the DCT coefficients, the first reference value, and the second reference value; and a mapping unit mapping the DCT coefficients to the adaptive quantization intervals.

Description

APPARATUS AND METHOD OF EMBEDDED QUANTIZATON FOR THE IMPROVED

SNR SCALBILTY

TECHNICAL FIELD The present invention relates to a quantization apparatus and method for providi ng improved Signal-to-Noise Ratio (SNR) scalability.

BACKGROUND ART

A conventional H.264-based quantization method assigns Discrete Cosine Trans form (DCT) coefficients to quantization intervals according to quantization parameters s et for respective frames. However, since the conventional H.264-based quantization method does not follow a distribution of DCT coefficients of each quantization interval g enerated by a video encoding method supporting picture quality scalability, encoding eff iciency is low.

DETAILED DESCRIPTION OF THE INVENTION

TECHNICAL PROBLEM

The present invention provides a quantization apparatus and method which obtai n a distribution of Discrete Cosine Transform (DCT) significant coefficients of the residu es of each SNR enhancement layer generated by a video encoder with improved pictur e-quality scalability, and assign DCT coefficients of the corresponding frame to an opti mal quantization interval, using Rate-Distortion (R-D) optimization, thereby providing hig h coding efficiency.

TECHNICAL SOLUTION

According to an aspect of the present invention, there is provided a quantization apparatus providing improved Signal-to-Noise Ratio (SNR) scalability, including: an R-D optimization unit performing Rate-Distortion (R-D) optimization based on a distribution of Discrete Cosine Transform (DCT) coefficients of each slice and calculating a first ref erence value and a second reference value respectively indicating a start point and an end point of DCT coefficients quantized to "0"; a quantization interval setting unit setting adaptive quantization intervals on the basis of a minimum value and a maximum value of the DCT coefficients, the first reference value, and the second reference value; and a mapping unit mapping the DCT coefficients to the adaptive quantization intervals.

According to another aspect of the present invention, there is provided an encod er providing SNR scalability, including: a quantization unit performing R-D optimization based on a distribution of DCT coefficients of each slice, calculating quantization coeffic ient values and reference values respectively indicating a start value and an end value of DCT coefficients quantized to "0", and performing quantization; and a dequantization unit performing dequantization based on average values of DCT coefficients of respecti ve intervals divided according to the reference values and the quantization coefficient v alues.

According to another aspect of the present invention, there is provided a codec p roviding improved SNR scalability, including: an R-D optimization unit performing R-D o ptimization based on a distribution of DCT coefficients of each slice and calculating a fir st reference value and a second reference value respectively indicating a start point an d an end point of DCT coefficients quantized to "0"; a quantization interval setting unit s etting adaptive quantization intervals based on a minimum value and a maximum value of the DCT coefficients, the first reference value, and the second reference value; a ma pping unit mapping the DCT coefficients to the adaptive quantization intervals; an entro py encoding unit adding, to a bit stream, values encoded based on average values of D CT coefficients of respective intervals divided according to the reference values; and a dequantization unit performing dequantization based on both the average value of the DCT coefficients and quantization coefficient values extracted from the bit stream.

According to another aspect of the present invention, there is provided a quantiz ation method providing improved SNR scalability including: performing R-D optimization based on a distribution of DCT coefficients of each slice and calculating a first referenc e value and a second reference value respectively indicating a start point and an end p oint of DCT coefficients quantized to "0"; setting adaptive quantization intervals on the b asis of a minimum value and a maximum value of the DCT coefficients, the first referen ce value, and the second reference value; and mapping the DCT coefficients to the ada ptive quantization intervals.

According to another aspect of the present invention, there is provided a coding method of providing improved SNR scalability, including: performing quantization after c alculating by performing R-D optimization on the basis of a distribution of DCT coefficie nts of each slice, calculating quantization coefficient values and reference values respe ctively indicating a start point and an end point of a range of DCT coefficients quantized to "0"; and performing dequantization on the basis of average values of DCT coefficien ts of each section divided based on the reference values and the quantization coefficien t values.

According to another aspect of the present invention, there is provided a comput er readable recording medium having embodied thereon a computer program for execu ting the method.

DESCRIPTION OF THE DRAWINGS

The above and other features and advantages of the present invention will beco me more apparent by describing in detail exemplary embodiments thereof with referenc e to the attached drawings in which:

FIG. 1 illustrates a hierarchical structure for providing picture-quality scalability in a video encoding method supporting picture-quality scalability;

FIG. 2 is a view for explaining cases in which a rounding artifact is generated in a JSVM progressive quantization method;

FIG. 3 is a view for more explaining a rounding artifact effect in more detail; FIG. 4 is a view for explaining non-significant quantization based on the JSVM 1. 0 standard;

FIG. 5 is a view for explaining significant quantization based on the JSVM 1.0 sta ndard;

FIG. 6 is a block diagram of a quantization apparatus for providing improved SN R scalability, according to an embodiment of the present invention; FIG. 7 is a graph showing a distribution of Discrete Cosine Transform (DCT) coef ficients;

FIG. 8 is a graph used to calculate reference values, according to an embodimen t of the present invention;

FIG. 9 is a view for explaining a process in which quantization intervals and recon struction values are calculated through Rate-Distortion (R-D) optimization from a distrib ution of DCT coefficients, according to an embodiment of the present invention;

FIG. 10 is a block diagram of an encoder including a quantization unit for providin g improved SNR scalability, according to an embodiment of the present invention; FIG. 11 is a flowchart illustrating a quantization method, according to an embodi ment of the present invention;

FIG. 12 is a flowchart illustrating a quantization and dequantization method, acco rding to an embodiment of the present invention; and FIGS. 13 through 19 are graphs showing effects obtained by the methods accord ing to the present invention.

BEST MODE

Hereinafter, the present invention will be described in detail by explaining preferr ed embodiments of the invention with reference to the attached drawings. Like referen ce numerals in the drawings denote like elements. In this specification, detailed descript ions related to functions or constructions well-known in the art will be omitted.

FIG. 1 is a block diagram illustrating a hierarchical structure for providing picture- quality scalability in a video encoding method supporting picture-quality scalability. Referring to FIG. 1 , a transformed image or an original image is input through the process of transformation, scaling, and quantization, and is generated as an encoded stream in a Signal-to-Noise Ratio (SNR) base layer. In order to generate an input ima ge of a SNR enhancement layer, the encoded stream of the SNR base layer is process ed by dequantization, descaling, inverse-transformation, and dequantization and thus is reconstructed as a low and a high image. A difference between the reconstructed im age and the original image is generated as an input image of a SNR enhancement laye r.

An enhancement encoding stream is generated in each enhancement layer, thro ugh the same method as that used in the SNR base layer, and transferred to a decoder . At this time, a quantization parameter used in each layer is a value obtained by subtr acting 6 from a quantization parameter used in the lower layer.

In this example, SNR scalability is provided by iterative quantization of the residu al signals computed between the original subband pictures and the reconstructed subb and pictures obtained after decoding the SNR base layer and previous SNR enhancem ent layers.

In the video encoding method supporting picture-quality scalability illustrated in Fl G. 1 , quantization is performed using the following Equation 1. =

qbits (a) sign(Zij) = sign(Wi_j) (b) (1)

In Equation 1 , Zy denotes a quantized coefficient, Wy denotes a DCT-transforme d result, MF denotes a multiplication factor, f denotes a rounding offset, and » denotes a right binary shift. In the H.264-based reference model software, f is 2^qblt/3 with resp ect to an intra block, and 2^qblt/6 with respect to an inter block.

In the video encoding method supporting picture-quality scalability, dequantizatio n is performed using the following Equation 2.

W^' _iJ = Z_ijV_iJ2^βoor(β'"⁶⁾ (2)

If quantization intervals based on the conventional H.264 standard are used, Equ ation 2 can be applied to an image encoding method supporting picture-quality scalabilit y. However, in the conventional H.264-based quantization method non-integer numbe rs are rounded to the nearest integer. Thus, it is impossible to extract the original non- integer number from the rounded integer is impossible, which causes an irreversible los s.

That is, the quantization and dequantization using Equations 1 and 2 do not folio w a distribution of Discrete Cosine Transform (DCT) coefficients of each slice, and cann ot obtain optimal quantization intervals and reconstruction values. Accordingly, encodi ng efficiency is low.

FIG. 2 is a diagram for explaining cases in which a rounding artifact is generated in a JSVM progressive quantization method. As illustrated in FIG. 2, in the JSVM progressive quantization method, due to the rounding artifact, quantization intervals are not perfectly embedded.

(a) in FIG. 2 illustrates a case where a quantization interval of a SNR enhanceme nt layer is perfectly embedded in a quantization interval of a SNR base layer. Meanwh ile, (b) and (c) in FIG. 2 illustrate cases where there is a difference between a quantizati on interval of a SNR base layer and a quantization interval of a SNR enhancement laye r due to a rounding artifact. That is, the difference between the coefficients of input re sidues of a current layer and dequantized coefficients of a current layer and the coeffici ents of input residues of the next enhancement layer is likely not to be coherent.

An effect of the rounding artifact will be described in detail with reference to FIG. 3, below. Referring to FIG. 3, a part of values encoded to "1" in a SNR base layer can be mapped to "-1" in a SNR enhancement layer, due to the rounding artifact effect. Also, a part of values encoded to "-1" in the SNR base layer can be mapped to "1" in the SNR enhancement layer, due to the round artifact effect.

FIG. 4 is a diagram for explaining non-significant quantization based on the conv entional JSVM 1.0 standard.

A case where a quantized DCT coefficient in a base layer is "0" is called "non-sig nificant". In this case, the DCT coefficient of the corresponding enhancement layer mu st be located in an area 411 illustrated in FIG. 4.

However, when quantization or dequantization is performed using Equations 1 an d 2, the DCT coefficient of the enhancement layer may be found in areas 412 and 413 as shown in FIG. 4. Also, due to the rounding artifact, it is difficult to correctly estimate intervals in which quantized DCT coefficients are located (420, 430).

FIG. 5 is a view for explaining significant quantization based on the conventional JSVM 1.0 standard. A case where a quantized DCT coefficient in a base layer is 1 is called "significan t". In this case, the DCT coefficient of the corresponding enhancement layer must be I ocated in an area 511 illustrated FIG. 5. However, like the case illustrated in FIG. 4, w hen quantization or dequantization is performed using Equations 1 and 2, the DCT coef ficient of the enhancement layer may be found in areas 512 and 513 as illustrated in Fl G. 5. Also, due to the rounding artifact, intervals in which quantized DCT coefficients a re located become narrower (520) or wider (530).

The above-described problems are solved in the method according to the presen t invention by obtaining a distribution of DCT coefficients of each quantization interval a nd assigning DCT coefficients of the corresponding frame to optimal quantization interv als using Rate-Distortion (R-D) optimization, thus making it possible to provide high codi ng efficiency.

FIG. 6 is a block diagram of a quantization apparatus 600 for providing improved SNR scalability, according to an embodiment of the present invention. The quantization apparatus 600 includes an R-D optimization unit 610, a quanti zation interval setting unit 620, and a mapping unit 630.

The R-D optimization unit 610 performs R-D optimization on the basis of a distrib ution of DCT coefficients of each slice and calculates a first reference value and a seco nd reference value respectively indicating a start point and an end point of a range of D CT coefficients quantized to "0".

In more detail, the R-D optimization unit 610 calculates a first reference value an d a second reference value which minimize a cost function J=D+ λ R. Here, D denote s an average distortion value, R denotes an average bit rate, and λ denotes a Lagrang e multiplier. The D and R values are expressed as follows.

^D = Σ ∑(*_k -δ_k)²p_k , and k=0 x_k z(a_k ,ot_M) 2

R =-∑Pk i°g(^p _k) _' i=0 where, P_k = n_k /N, p_Xι = n_Xι /N, and

^δk = ∑^xi ^lnk

*,e(α*.«*+i)

Here, N denotes the total number of Wy, and nι_< denotes the number of Wy in a ra nge [α _k> α _k+i]- A process for calculating the first reference value and the second refe rence value will be described in detail later with reference to FIGS. 7 and 8.

The quantization interval setting unit 620 sets adaptive quantization intervals, on the basis of minimum and maximum values of the DCT coefficients and the first and se cond reference values calculated by the R-D optimization unit 610. In this case, if the minimum value of the DCT coefficients is cto, the maximum value of the DCT coefficient s is c(₃, the first reference value is α-i, and the second reference value is α₂, the adaptive quantization interval can be obtained as illustrated in (b) of FIG. 9.

After the first reference value O₁ and the second reference value α₂ are calculate d, three intervals (that is, [α₀, αi ], [α-i, α₂ ], and [α₂, α₃]) of Wy quantized to "1", "0", and " -1" are determined. That is, reconstruction values δ i, δ ₂, and δ ₃ of the intervals [α o, Ct₁ ], [α-i, α₂ ], and [α₂, α₃] are average values of Wy in the respective intervals [α₀, αi ],

[αi, α₂ ], and [Ci₂, α₃].

The mapping unit 630 maps the DCT coefficients to the adaptive quantization int ervals, thereby performing quantization. In detail, coefficients from the minimum value do of the DCT coefficients to the first reference value cti are mapped to "-1", coefficient s from the second reference value α₂ to the maximum value α₃ of the DCT coefficients are mapped to "1", and the remaining coefficients are mapped to "0".

CC₂

CC₃

FIG. 7 is a graph showing a distribution of the DCT coefficients Wy. In FIG. 7, an x axis corresponds to values of the DCT coefficients Wy, and a y axi s corresponds to the number of the DCT coefficients Wy.

FIG. 8 is a graph used to calculate reference values, according to an embodime nt of the present invention.

Referring to FIGS. 6, 7, and 8, the R-D optimization unit 610 searches for a distri bution histogram of DCT coefficients Wy of enhancement layer of each fine granularity s calability (FGS). Then, the R-D optimization unit 610 calculates a first reference value αi and a second reference value α₂ so that R≤ RAVC is satisfied and a cost function J=D +Λ R is minimized, based on a distribution of DCT coefficients of each slice. Here, RA vc means entropy obtained by reconstruction values and step sizes of a conventional q uantization method.

The average distortion value D and the average bit rate R are obtained by varyin g the first reference value αi from 0 to the minimum value αo of the DCT coefficients an d varying the second reference value α₂ from 0 to the maximum value α₃ of the DCT co efficients.

FIG. 8 shows R and D values obtained according to the minimum value αo of the DCT coefficients and the first reference value α-i. In FIG. 8, values which minimize the cost function value J are connected by a solid line. A point which satisfies R≤ RAVC of the values is a point 810. Accordingly, the first reference value αi and the minimum v alue α₀ of DCT coefficients indicate R and D values corresponding to the point 810.

By calculating the first reference value αi and the second reference value α₂ thro ugh R-D optimization using the method according to the present invention, a value f whi ch decides the size of a dead-zone varies instead of being fixed as in a conventional m ethod. That is, f=2^qblt+ CH when Wy is a positive number and f = 2^qblt - α₂ when Wy is a negative number, wherein qbit=15+floor(Q^p/6).

FIG. 9 is a diagram for explaining an example in which reconstruction values and quantization intervals which minimize a cost function J= D+ λ R are obtained by perfo rming R-D optimization, from the distribution of the DCT coefficients.

As illustrated in (a) of FIG. 9, a conventional JSVM quantizer is manufactured un der an assumption that DCT coefficients exist in an area 910. However, actually, DCT coefficients of SNR enhancement layers can exist only in an area 920.

Accordingly, the present invention provides a quantization apparatus, which is ca pable of adaptively setting quantization intervals, considering only parts in which DCT c oefficients of SNR enhancement layers actually exist, thereby achieving higher encodin g efficiency than the conventional JSVM quantization apparatus.

FIG. 10 is a block diagram of an encoder 1000 including a quantization unit for pr oviding improved SNR scalability, according to an embodiment of the present invention.

The encoder 1000 includes a quantization unit 1010, a dequantization unit 1020, and an entropy coding unit 1030.

The quantization unit 1010 performs R-D optimization based on a distribution of DCT coefficients of each slice, calculates quantization coefficient values and reference values indicating a start value and an end value of DCT coefficients quantized to "0", an d performs quantization.

In this case, the function and technical concept of the quantization unit 1010 is th e same as that of the quantization apparatus 600 described above with reference to Fl G. 6, and therefore, a detailed description thereof will be omitted. The entropy coding

unit 1030 encodes Y = ^ — without encoding δj in order to reduce the numb

er of bits to be transmitted. The dequantization unit 1020 performs dequantization using Equation 3, on the b asis of values Zy encoded by the quantization unit 1010 and average values δo, δ-i, and δ₂ of DCT coefficients of respective intervals, wherein δ₀ is an average value of DCT co efficients in the interval [α₀, α-i], δi is an average value of DCT coefficients in the interva I [α-i, α₂], and δ₂ is an average value of DCT coefficients in the interval [α₂, 0:₃].

Wi/ =(Zϋ +Y_k)V_ij2^{floor(QP/6) '} (3)

In this case, Y = , and k are decided by the following Equation 4.

A codec (not illustrated) for providing improved SNR scalability using the quantiz ation method according to the present invention, includes an encoder and a decoder. The encoder includes an R-D optimization unit, a quantization interval setting unit, a ma pping unit, and an entropy encoder. The decoder includes a dequantization unit.

A technical concept for quantizing DCT coefficients through the R-D optimization unit, the quantization interval setting unit, and the mapping unit is described above with reference to FIG. 6, and therefore, a detailed description thereof is omitted.

The entropy encoder adds, to a bit stream, compressed values

2^l5+βoc"^'<-^Q > Y - _jr. — of the DCT coefficient average values δo, δi, and δ₂ of the respectiv

e sections divided based on the first reference value and the second reference value in the adaptive quantization intervals set by the quantization interval setting unit, with resp ect to the DCT coefficients quantized through the mapping unit. The entropy encoder

encodes and transmits the compression values F = without encoding δj

, in order to reduce the number of bits to be transmitted. In the decoder, the dequantization unit performs dequantization based on the qu antization coefficients Zy and the DCT coefficient average values δo, δi, and 6₂ extracte d from the bit stream.

FIG. 11 is a flowchart illustrating a quantization method, according to an embodi ment of the present invention.

Referring to FIGS. 10 and 11 , a distribution of DCT coefficients of each slice rece ived through the quantization unit 1010 of the encoder 1000 is obtained (operation S11 10). Then, R-D optimization is performed based on the obtained distribution of the DC T coefficients, so that a first reference value αi and a second reference value α₂ indicati ng a start value and an end value of a range of DCT coefficients quantized to "0" are cal culated (operation S1120).

Then, quantization interval setting for setting adaptive quantization intervals base d on a minimum value Wj_j__min and a maximum value Wjj__maxof the DCT coefficients, the first reference value α-i, and the second reference value 0₂, is performed (operation S11 30). Thereafter, mapping for mapping the DCT coefficients to the adaptive quantizatio n intervals is performed (operation S1140). In operation S1140, coefficients from the minimum value Wi_j__min of the DCT coefficients to the first reference value αi are mappe d to "-1", coefficients from the second reference value α₂ to the maximum value Wjj__maχ of the DCT coefficients are mapped to "1", and the remaining coefficients are mapped t 0 "0".

FIG. 12 is a flowchart illustrating a quantization and dequantization method, acco rding to an embodiment of the present invention.

A coding method for providing SNR scalability includes quantization by an encod er and dequantization by a decoder. First, R-D optimization is performed based on a distribution of DCT coefficients of each slice, quantization coefficient values and reference values indicating a start value and an end value of a range of DCT coefficients quantized to "0" are calculated, and th en quantization is performed.

In more detail, R-D optimization is performed based on a distribution of DCT coef ficients of each slice, R-D optimization (operation S1220) for calculating a first referenc e value and a second reference value indicating a start value and an end value of a ran ge of DCT coefficients to be quantized to "0" is performed, and then quantization interva

I setting (operation S1230) for setting adaptive quantization intervals on the basis of the maximum and minimum values of the DCT coefficients and the first and second refere nee values calculated in the operation S1220 is performed.

Coefficients from the minimum value of the DCT coefficients to the first reference value are mapped to "-1", coefficients from the second reference value to the maximu m value of the DCT coefficients are mapped to "1", and the remaining coefficients are mapped to "0", thereby performing quantization (operation S1240).

Thereafter, entropy encoding (operation S 1250) for adding to a bit stream values encoded on the basis of average values of DCT coefficients of respective intervals divi ded according to the reference values, is performed, and the bit stream is transferred to the decoder.

In the decoder, dequantization is performed (operation S1260). That is, reconst

ruction values δo, δi, and 62 are calculated on the basis of Y = — 2 ,l5_«+_,fl„o_,o_„r,(Q _P ^p _l'_i ⁶ —) informati

on included in the bit stream, and dequantization is performed using Equations 3 and 4 (operation S 1260). FIGS. 13 through 19 are graphs illustrating results obtained using the methods a ccording to the present invention.

FIGS. 13 through 19 show results when a FGS layer is stacked on the correspon ding layer in image format. Frame rate conditions are denoted above each graph.

In FIGS. 13 through 19, left and lower points are rate distortion points of a base I ayer, and right and upper points are rate distortion points of a first FGS layer.

In FIGS. 17, 18, and 19, the proposed method has characteristics almost identica I to the conventional method. However, in FIGS. 13, 14, and 15, the proposed method has performance improved by about 0.1 dB, by about 1 dB, and by about 0.8 dB, resp ectively, compared to the conventional method. The present invention can also be embodied as computer readable codes on a c omputer readable recording medium. The computer readable recording medium is an y data storage device that can store data which can be thereafter read by a computer s ystem. Examples of the computer readable recording medium include read-only mem ory (ROM), random-access memory (RAM), CD-ROMs, magnetic tapes, floppy disks, 0 ptical data storage devices, and carrier waves (such as data transmission through the I nternet). The computer readable recording medium can also be distributed over netwo rk coupled computer systems so that the computer readable code is stored and execute d in a distributed fashion.

While the present invention has been particularly shown and described with refer ence to exemplary embodiments thereof, it will be understood by those of ordinary skill i n the art that various changes in form and details may be made therein without departin g from the spirit and scope of the present invention as defined by the following claims.

INDUSTRIAL APPLICABILITY

As described above, according to the present invention, by optimally calculating quantization intervals and reconstruction values through a distribution of DCT coefficien ts of each frame and performing encoding and decoding, when DCT coefficients of eac h SNR enhancement layer are quantized in scalable video coding, high coding efficienc y can be achieved.

Claims

1. A quantization apparatus providing improved Signal-to-Noise Ratio (SNR) scalab ility, comprising: an R-D optimization unit performing Rate-Distortion (R-D) optimization based on a distribution of Discrete Cosine Transform (DCT) coefficients of each slice and calculat ing a first reference value and a second reference value respectively indicating a start p oint and an end point of DCT coefficients quantized to "0"; a quantization interval setting unit setting adaptive quantization intervals on the b asis of a minimum value and a maximum value of the DCT coefficients, the first referen ce value, and the second reference value; and a mapping unit mapping the DCT coefficients to the adaptive quantization interva IS.

2.

The quantization apparatus of claim 1 , wherein the R-D optimization unit calculat es the first reference value and the second reference value so that a cost function J=D+ λ R is minimized, where D denotes an average distortion value, R denotes an average bit rate, and λ denotes a Lagrange multiplier.

3.

The quantiziation apparatus of claim 1 , wherein the mapping unit maps coefficie nts from the minimum value of the DCT coefficients to the first reference value to "-1", c oefficients from the second reference value to the maximum value of the DCT coefficie nts to "1", and the remaining coefficients to "0".

4. An encoder providing SNR scalability, comprising: a quantization unit performing R-D optimization based on a distribution of DCT c oefficients of each slice, calculating quantization coefficient values and reference value s respectively indicating a start value and an end value of DCT coefficients quantized to "0", and performing quantization; and a dequantization unit performing dequantization based on average values of DC T coefficients of respective intervals divided according to the reference values and the q uantization coefficient values.

5. The encoder of claim 4, wherein the quantization unit comprises: an R-D optimization unit performing R-D optimization based on a distribution of D

CT coefficients of each slice and calculating a first reference value and a second refere nee value respectively indicating a start point and an end point of DCT coefficients quan tized to 'O"; a quantization interval setting unit setting adaptive quantization intervals based o n a minimum value and a maximum value of the DCT coefficients, the first reference val ue, and the second reference value; and a mapping unit mapping the DCT coefficients to the adaptive quantization interva

Is.

6.

The encoder of claim 5, wherein the R-D optimization unit calculates the first refe rence value and the second reference value so that a cost function J=D+λ R is minimiz ed, where D denotes an average distortion value, R denotes an average bit rate, and λ denotes a Lagrange multiplier.

7. The encoder of claim 5, wherein the mapping unit maps coefficients from the min imum value of the DCT coefficients to the first reference value to "-1", coefficients from t he second reference value to the maximum value of the DCT coefficients to "1", and the remaining coefficients to "0".

8.

The encoder of claim 4, further comprising an entropy encoding unit adding, to a bit stream, values encoded based on average values of DCT coefficients of respective i ntervals divided according to the reference values.

9. The encoder of claim 8, wherein the average values of the DCT coefficients of th e respective intervals divided according to the reference values are σ , and compresse

d values of the average values are Y, Y

10. A codec providing improved SNR scalability, comprising: an R-D optimization unit performing R-D optimization based on a distribution of D

CT coefficients of each slice and calculating a first reference value and a second refere nee value respectiveiy indicating a start point and an end point of DCT coefficients quan tized to "0"; a quantization interval setting unit setting adaptive quantization intervals based o n a minimum value and a maximum value of the DCT coefficients, the first reference val ue, and the second reference value; a mapping unit mapping the DCT coefficients to the adaptive quantization interva Is; an entropy encoding unit adding, to a bit stream, values encoded based on aver age values of DCT coefficients of respective intervals divided according to the referenc e values; and a dequantization unit performing dequantization based on both the average valu e of the DCT coefficients and quantization coefficient values extracted from the bit strea m.

11.

A dequantization apparatus, wherein quantization coefficient values and average values of DCT coefficients of respective intervals divided on the basis of a start point a nd an end point of a range of DCT coefficients quantized to "0", in quantization intervals set based on a distribution of the DCT coefficients, are extracted from an encoded bit stream, and dequantization is performed.

12.

The dequantization apparatus of claim 11 , wherein the quantization intervals are set on the basis of a start point and an end point of a range of DCT coefficients quantiz ed to "0" extracted by performing R-D optimization based on a maximum value and a mi nimum value of the DCT coefficients and the distribution of the DCT coefficients.

13.

The dequantization apparatus of claim 12, wherein the R-D optimization is to cal culate a first reference value and a second reference value so that a cost function J=D+ λ R is minimized, where D denotes an average distortion value, R denotes an average bit rate, and λ denotes a Lagrange multiplier.

14. A decoder providing improved SNR scalability comprising: a receiver receiving an encoded bit stream; an extractor extracting a DCT coefficient average value and a quantization coeffi cient value from the received bit stream; and a dequantization unit performing dequantization on the basis of the DCT coefficie nt average value and the quantization coefficient value.

15.

The decoder of claim 14, wherein the DCT coefficient average value of each sect ion of adaptive quantization intervals is an average value of DCT coefficient present in a corresponding section, and the sections of the adaptive quantization interval are set u sing a maximum value and a minimum value of the DCT coefficients of the residues of each SNR enhancement and a start point and an end point of a range of DCT coefficie nts to be quantized to "0", the start point and the end point being extracted by performin g R-D optimization.

16. The decoder of claim 15, wherein the R-D optimization is to calculate a first refer ence value and a second reference value so that a cost function J=D+λ R is minimized, where D denotes an average distortion value, R denotes an average bit rate, and λ d enotes a Lagrange multiplier.

17. A quantization method providing improved SNR scalability comprising: performing R-D optimization based on a distortion of DCT coefficients of each sli ce and calculating a first reference value and a second reference value respectively indi eating a start point and an end point of DCT coefficients quantized to "0"; setting adaptive quantization intervals on the basis of a minimum value and a ma ximum value of the DCT coefficients, the first reference value, and the second referenc e value; and mapping the DCT coefficients to the adaptive quantization intervals.

18.

The quantization method of claim 17, wherein the R-D optimization is to calculat e the first reference value and the second reference value so that a cost function J=D+ λ R is minimized, where D denotes an average distortion value, R denotes an average bit rate, and λ denotes a Lagrange multiplier.

19. The quantization method of claim 17, wherein, in the mapping, coefficients from t he minimum value of the DCT coefficients to the first reference value are mapped to "-1 ", coefficients from the second reference value to the maximum value of the DCT coeffi cients are mapped to "1", and the remaining coefficients are mapped to "0", respectively

20. A coding method of providing improved SNR scalability, comprising: performing quantization after calculating by performing R-D optimization on the b asis of a distribution of DCT coefficients of each slice, calculating quantization coefficie nt values and reference values respectively indicating a start point and an end point of a range of DCT coefficients quantized to "0"; and performing dequantization on the basis of average values of DCT coefficients of each section divided based on the reference values and the quantization coefficient val ues.

21.

The coding method of claim 20, wherein the performing of the quantization comp rises: performing R-D optimization based on a distortion of DCT coefficients of each sli ce and calculating a first reference value and a second reference value respectively indi eating a start point and an end point of DCT coefficients quantized to "0"; setting adaptive quantization intervals on the basis of a minimum value and a ma ximum value of the DCT coefficients, the first reference value, and the second referenc e value; and mapping the DCT coefficients to the adaptive quantization intervals.

22.

The coding method of claim 21 , wherein the R-D optimization is to calculate the f irst reference value and the second reference value so that a cost function J=D+λ R is minimized, where D denotes an average distortion value, R denotes an average bit rate , and λ denotes a Lagrange multiplier.

23.

The coding method of claim 21 , wherein, in the mapping, coefficients from a mini mum value of the DCT coefficients to the first reference value are mapped to "-1", coeffi cients from the second reference value to a maximum value of the DCT coefficients are mapped to "1", and the remaining coefficients are mapped to "0".

24.

The coding method of claim 20, further comprising entropy encoding for adding, t o a bit stream, values encoded based on average values of DCT coefficients of each of sections divided according to the first and second reference values

25.

The coding method of claim 24, wherein the average values of the DCT coefficie nts of each of sections divided according to the first and second reference values are

σ , and compressed values of the average values are Y, Y

26.

A dequantization method providing improved SNR scalability, comprising: extract ing a quantization coefficient value and a DCT coefficient average value of each of secti ons divided on the basis of a start point and an end point of a range of DCT coefficients quantized to "0", in an adaptive quantization interval set based on a distribution of the DCT coefficients, from an encoded bit stream.

27.

The dequantization method of claim 26, wherein the adaptive quantization interv al is set using a maximum value and a minimum value of the DCT coefficients of the re sidues of each SNR enhancement and a start point and an end point of a range of DCT coefficients to be quantized to "0" based on, the start point and the end point being ext racted by performing R-D optimization.

28.

The dequantization method of claim 27, wherein the R-D optimization is to calcul ate a first reference value and a second reference value so that a cost function J=D+λ R is minimized, where D denotes an average distortion value, R denotes an average bit rate, and λ denotes a Lagrange multiplier.

29. A decoding method of providing improved SNR scalability comprising: receiving an encoded bit stream; extracting a DCT coefficient average value and a quantization coefficient value fr om the bit stream; and performing dequantization based on the DCT coefficient average value and the q uantization coefficient value.

30.

The decoding method of claim 29, wherein the DCT coefficient average value is set using a maximum value and a minimum value of the DCT coefficients of the residue s of each SNR enhancement and a start point and an end point of a range of DCT coeff icients to be quantized to "0", the start point and the end point being extracted by perfor ming R-D optimization.

31. The decoding method of claim 30, wherein the R-D optimization is performed by calculating a first reference value and a second reference value so that a cost function J=D+λ R is minimized, where D denotes an average distortion value, R denotes an ave rage bit rate, and λ denotes a Lagrange multiplier.

32.

A computer readable recording medium having embodied thereon a computer pr ogram for executing any one of claims 17 through 31.